Add ifdef pragma or delete unneeded pragma

aidge_export_cpp/kernels/add.hpp

@@ -55,12 +55,10 @@ void add_forward (
     if (contiguousidx > 0) {
         stride_post0[contiguousidx  - 1] = 1;
         stride_post1[contiguousidx  - 1] = 1;
-        #pragma omp parallel for
         for (int i = contiguousidx -2; i != -1; --i) {
             stride_post0[i] = stride_post0[i+1]*ndim_a[i+1];
             stride_post1[i] = stride_post1[i+1]*ndim_b[i+1];
         }
-        #pragma omp parallel for
         for (int i = 0; i < contiguousidx ; ++i) {
             stride_step0[i] = (ndim_a[i] == 1) ? 1 - stride_post0[i] : 1;
             stride_step1[i] = (ndim_b[i] == 1) ? 1 - stride_post1[i] : 1;

aidge_export_cpp/kernels/batchnorm.hpp

@@ -25,15 +25,21 @@ void batchnorm_forward (
 {
 
     int featureMapSize = OUTPUTS_HEIGHT * OUTPUTS_WIDTH;
+#ifdef _OPENMP
     #pragma omp parallel for
+#endif
     for (int ch = 0; ch < NB_OUTPUTS; ++ch) {
         int ioIndex = ch * featureMapSize;
+#ifdef _OPENMP
         #pragma omp parallel for
+#endif
         for (int i = ioIndex; i < ioIndex + featureMapSize; i++) {
             outputs[i] = biases[ch];
         }
         float var = sqrt(variances[ch] + epsilon);
+#ifdef _OPENMP
         #pragma omp parallel for
+#endif
         for (int feature = 0; feature < featureMapSize; ++feature) {
             outputs[ioIndex + feature] += (scales[ch] * (inputs[ioIndex + feature] - means[ch]) / var);
         }

aidge_export_cpp/kernels/convolution_groups.hpp

@@ -6,7 +6,6 @@
 #include "network/utils.hpp"
 #include "kernels/macs.hpp"
 #include "kernels/activation.hpp"
-#include <omp.h>
 
 // Weights index en NHWC
 constexpr int inds_pos(int n, int c, int h, int w, int N, int C, int H, int W) {
@@ -52,19 +51,22 @@ void convolution_forward(
     
     int c_in_g = NB_CHANNELS / GROUPS;
     int c_out_g = NB_OUTPUTS / GROUPS;
-    #pragma omp parallel for
+#ifdef _OPENMP
+    #pragma omp parallel for collapse(3)
+#endif
     for (int oc = 0; oc < NB_OUTPUTS; oc++) {
-    	int g_oc = oc / c_out_g;
-        #pragma omp parallel for
     	for (int i = 0; i < OUT_HEIGHT; ++i) {
-            #pragma omp parallel for
     	    for (int j = 0; j < OUT_WIDTH; ++j) {
+                int g_oc = oc / c_out_g;
     	        Output_T value = biases[oc];
-                #pragma omp parallel for
     	        for (int ic = g_oc * c_in_g; ic < (g_oc + 1) * c_in_g; ++ic) {
+#ifdef _OPENMP
                     #pragma omp parallel for
+#endif
     	            for (int m = 0; m < KERNEL_HEIGHT; ++m) {
+#ifdef _OPENMP
                         #pragma omp parallel for
+#endif
     	                for (int n = 0; n < KERNEL_WIDTH; ++n) {
     	                    int i_p = i * STRIDE_X - PADDING_X + m * DILATION_X;
                             int j_p = j * STRIDE_Y - PADDING_Y + n * DILATION_Y;

aidge_export_cpp/kernels/erf.hpp

@@ -19,8 +19,9 @@ void erf_forward (
     double a5 =  1.061405429;
     double p  =  0.3275911;
 
-
+#ifdef _OPENMP
 #pragma omp parallel for
+#endif 
     for (int i = 0; i < _NB_ELTS; ++i) {
         int sign = 1;
         if (inputs[i] < 0)

aidge_export_cpp/kernels/mul.hpp

@@ -54,12 +54,10 @@ void mul_forward (
     if (contiguousidx > 0) {
         stride_post0[contiguousidx  - 1] = 1;
         stride_post1[contiguousidx  - 1] = 1;
-        #pragma omp parallel for
         for (int i = contiguousidx -2; i != -1; --i) {
             stride_post0[i] = stride_post0[i+1]*ndim_a[i+1];
             stride_post1[i] = stride_post1[i+1]*ndim_b[i+1];
         }
-        #pragma omp parallel for
         for (int i = 0; i < contiguousidx ; ++i) {
             stride_step0[i] = (ndim_a[i] == 1) ? 1 - stride_post0[i] : 1;
             stride_step1[i] = (ndim_b[i] == 1) ? 1 - stride_post1[i] : 1;